What do statisticians research?

[@John-qo9hw]

Wow, you helped me get rid of a lot of the black box I had in my mind when I heard the word statistics. Thank you.

[@qwerty11111122]

He hasn't done a map on statistics yet?!

[@marcbollee8052]

Super interesting! I'd love to see a video about some of the more obscure topics you mentioned. Very high quality content, thank you!

[@sanketkhatavkar7531]

Thank you for thoroughly deep diving into stats and presenting to us these interesting informational topics.

[@TheMegaDTGT48]

Hey man. Thank You for this video. I’m planning on doing the PhD in stat so this will be very helpful 🤗 It’s funny as for me the Graphical Models are the most interesting part of statistics. I’m working in this subfield for over a year now haha

[@mx20771]

I’m a community college student soon going to a larger university and will be majoring in statistics (and potentially biostatistics as I want to have a positive impact in the space). I just want to say I love your videos and your ability to foster intuition. A true goldmine, thank you! p.s. If you have any advice for the path of a thoughtful and strong statistician/biostatistician I’d be grateful

[@mike74h]

Statistics is a tool. Whether you use it to have a positive impact is independent of which branch is your specialty, or whether you have one. The corollary is that Twain's quote can apply, regardless. It's great to want to make a positive contribution, though.

[@illbeback12]

Refreshing to see a video that mentions clustering and dynamic clustering in the era of the holy GPT. Back in the day (3 years ago) I was implementing Dynamic Topic Models (David Blei) in uni, and this made me nostalgic!

[@YuweiCheng-yy3uk]

Really enjoy all your videos. Very unique educational channel on statistics. Keep uploading!

[@programmingwithian]

I'm taking a class on Time Series Analysis next semester. Quite excited for it!

[@twentytwentyeight]

When you said “we’re using statistics to explain statistics”, I instantly subbed to ❤ wish I’d found you sooner (shakes fist vigorously at yt algo)

[@Kram1032]

Graphs (in the sense of graphical models) are hugely important as they basically can capture any type of pairwise relationship you might want. Specializing to specific types of graphs you can capture decision trees, causal orders, processes, various topological constraints, or many other notions of interest. And nearly anything can be framed in terms of graphs. So it's no wonder that's a huge area in statistics. (There are extensions that allow you to capture higher order relationships that consider three or more objects at a time. Multi graphs, hypergraphs, Abstract Simplical Complexes, Simplical Sets (different generalizations of graphs). But you can get really really far before you actually need those. Almost anything can be captured with just pairwise relationships)

[@very-normal]

Thanks! It turned out that some of my peers in my program did some graphical stuff, and your comment revived some memories to me. I feel like I got the bigger idea of the tool from how you phrased it, appreciate it!

[@tamirmashbat3147]

Stunned by what my Statistics is from what i've learned so far

[@matteogirelli1023]

I've studied in Bologna and confirm that it's a very good uni, especially in stats and applied stats

[@sharks1349]

Thank you for these videos. Big Help

[@daniellopez-dq1yz]

How did you become a biostatistician? Unfortunately where I live there almost no Biostatistics master programs

[@MusicEngineeer]

I think I recognize this cosine distance as something like "one minus correlation" or "one minus covariance" or something like that. Right? ...and I guess that it never gets negative is only because the x_i, y_i are constrained to be positive?

[@bin4ry_d3struct0r]

Would throwing a dart at a dartboard 100x and then choosing the one time you hit the bull's eye (while discarding the other 99 trials from the results) to show an accuracy rate of 100% be an example of p-hacking? This is the one I always use (afaik, I came up with it myself) and never bothered to confirm with anyone if I'm using it correctly.

[@very-normal]

Yeah that’s pretty good! If I were to add anything, it would just be a small detail that your darts land all across the board, as an indication that there’s no skill involved. That one errant bullseye is an indication of skill in a scenario when it was all luck. Great metaphor!

[@pichirisu]

Would you please consider doing a video going over the the amount/type/etc. of coding that is required by statisticians? I'd love to be a statistician and am doing undergrad for it atm with basic statistical packages which is totally fine and a lot of fun, but if I have to get more into data science-type work then I'm absolutely checked out cause that kind of coding instantly bores me. It would be highly appreciated, thanks.

[@very-normal]

Yeah! I’ve been pondering how to incorporate more actionable code-based videos into the channel and I have a few ideas. Keep your eyes peeled!

[@jeffreychandler8418]

I would say generally, most coding falls outside of the statistical analysis, especially at the undergrad level. So stuff like exploration, cleaning, etc take much more space, but that changes a bit as you get more advanced into statistics and are trying to, effectively, write the packages you are using now. So for example, just with a small part of my project, I used the "Earth Mover's Distance" and I'm simply computing it in a univariate case. This is done efficiently via the hungarian algorithm, which in my script makes it about 30 lines for two time periods. For fun I wanted to calculate the algorithm used for multidimensional EMD, about 100 lines of clear code, then I wanted to write the regularized form to improve time efficiency, 200 lines but WAY faster. Given what you say about yourself, you will love more advanced statistics. There's always a place for the mean(df), but there is ALWAYS value in linear algebra, linear programming (see "the art of linear programming" by Tom S, a gorgeous description of linear programming and the sinkhorn algorithm), pure maths, and computer science. I'm always learning interesting new ways to make the statistics more efficient, more flexible, etc.

[@pichirisu]

@@jeffreychandler8418 Thank you for your insight.

[@brandonheaton6197]

Hanging around biostatisticians, surely you have heard of Neo4J and knowledge graphs. Graphs (vertices and edges) are used to represent protein/protein interactions or metabolic pathways and their interconnections. They can also represent other levels of interaction, say cells in an organism or species in an ecosystem

[@very-normal]

That makes sense. I guess my ignorance comes from the fact that my own research deals with experimental design, so I don’t get many chances to rub shoulders with this type of research. I’ve definitely seen omics type of stats papers, but it didn’t occur to me that that’s the kind of model you’d want to use. Thanks for the examples!

[@jeffreychandler8418]

graph theory is also making it's way into survival/population analysis. Sort of started with Leslie projection matrices (very simple 1:1:1... graphs) and now with community calculations we're starting to see more usage of bayesian networks and the like to assess communities. It's mostly relegated to underappreciated grad student work, but I think with persistence it will grow

[@samuelm4719]

Cool! Time series was most of my advanced econometrics class, so I was surprised to hear that in other fields it's not as big a deal. But that makes sense I guess.

[@very-normal]

That’s really interesting, can I ask for your recommendation about what a good intro-level econometrics textbook would be? I’ve always been curious to see how stuff is covered from that perspective

[@twentytwentyeight]

Also would love any book recommendations!

[@samuelm4719]

@@very-normal I learned from Principles of Econometrics 5th ed. by Carter Hill, William Griffiths and Guay Lim. It covers all of the basics, and most of the advanced stuff, but maybe not all of the most sophisticated modelling strategies that a PhD would need to know. For the most part I felt like it was a pretty good text, but there are some things that it doesn't always explain in the depth I need to answer particularly tricky modelling issues.

[@matteogirelli1023]

If you want a reference for time series econometrics, see Hamilton "Time series analysis". Need quite a bit of math though.

[@BorrWick]

Why wouldn't they use a probabilistic topic model. Blei et al 2006 Dynamic Topic Models would be a perfect fit for the task

[@wanfuse]

Perhaps I missed it but I don’t see a link to the paper?

[@very-normal]

Oops, forgot to put it in! It's in the description, and here it is for convenience: arxiv.org/abs/1709.03563

[@overanalyzed5258]

im sure youve answered this before but where are you doing your phd?

[@very-normal]

lol actually I haven’t before. While I’m a student, I’ll keep my exact school under wraps until I graduate in case my department finds out, but I will say that I’m in Southern California! For biostat stuff, your choices are really isolated to either Cali or the East Coast

[@overanalyzed5258]

@@very-normal gotcha. Would your department not like it if they knew you did this? I'm a first year at Columbia planning on studying data science, maybe doing ai research or bio computing or even epidemiology, and thus channel is helping me see the beauty in stats

[@very-normal]

lol nah, I know faculty like to gossip so I’m just being private about it. That’s cool! I did my MS in biostats at Columbia. I’m glad the channel could be helpful, let me know if there’s anything that you’d like me to cover and I’ll try to figure out how to get the info out there

[@kristianwichmann9996]

The article seems like a nice review of a lot of "classical" NLP methods. Clearly pre-LLM-revolution.

[@riccardopucci3165]

Me watching my professors appear on screen all of a sudden 👁👄👁

[@pietroanedda2894]

Studi a Bologna? Come ti sembrano queste prof? Chiedo perché sto pensando di fare la magistrale là

[@josealbertofernandez4593]

Bro what's your uni ? i wanna meet you